<![CDATA[This study classifies YouTube users’ sentiment toward the issue of the Indonesian House of Representatives’ allowance increase using a TF-IDF-based Support Vector Machine (SVM). The dataset consists of 26,074 cleaned comments, with a class distribution of 74.61 percent negative and 25.39 percent positive. Initial labeling was performed using a lexicon-based approach and was validated on a limited scale through manual annotation of 150 samples, with a Cohen’s Kappa value of 0.8298. The Linear Kernel SVM model with `class_weight=‘balanced’` achieved an accuracy of 97.10 percent, an F1-score of 98.05 percent for the negative class, and 94.40 percent for the positive class. Using `class_weight=‘balanced’` did not improve all metrics, but it increased the positive class recall from 95.47 percent to 96.15 percent. The comparison results show that the Linear Kernel outperforms the RBF Kernel on high-dimensional data. Despite the high classification performance, 151 misclassified data points were found, influenced by contextual ambiguity, sarcasm, slang, and lexical coverage limitations. Further research is recommended to expand manual validation and utilize contextual semantic features.]]>
Copyrights © 2026
